It has the attempt reference from our documented handshake and a completion result; the display adds on a high transition, including its first read after reconnect, without consulting the stored result

I'd replay the saved before-and-after records through that display code first. You don't need a new inspection to show that a reconnect invents an extra count, and you can preserve the original result while testing it.

Is the stored result durable across application restart, Farah? A reconnect fix that only remembers this session will leave the same problem waiting behind a restart.

Results persist, processed observations don't; developer is making the report derive from the result records instead of incrementing on the initial high, and the replay currently reproduces the extra count

Check what the result actually means before naming that derived number accepted. Completed inspection can include a reject or something still awaiting disposition; I wouldn't have the reporting fix quietly decide that for quality.

You're right, Jasper, it was labelled accepted but counted completed attempts; quality is reconciling the identified plates and their dispositions, while developer keeps attempt results separate from the item summary

That is a bigger correction than the reconnect arithmetic. I'd retain one plate with a failed attempt followed by a reviewed pass in the tests, so a tidy new total doesn't come from deleting the uncomfortable attempt.

Replay now keeps the failed attempt and counts its reviewed accepted plate once; repeated reads and application restart leave the derived total unchanged, but two older results lack a plate link and remain unassigned for quality rather than becoming two more accepted items

Keep those unassigned results visible with a review owner. Also replay a completion first recovered after the outage; it must be reconciled once, not discarded merely because it wasn't already local.